2 * Copyright (c) 2009, Microsoft Corporation.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, see <http://www.gnu.org/licenses/>.
17 * Haiyang Zhang <haiyangz@microsoft.com>
18 * Hank Janssen <hjanssen@microsoft.com>
20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
22 #include <linux/init.h>
23 #include <linux/atomic.h>
24 #include <linux/module.h>
25 #include <linux/highmem.h>
26 #include <linux/device.h>
28 #include <linux/delay.h>
29 #include <linux/netdevice.h>
30 #include <linux/inetdevice.h>
31 #include <linux/etherdevice.h>
32 #include <linux/skbuff.h>
33 #include <linux/if_vlan.h>
35 #include <linux/slab.h>
37 #include <net/route.h>
39 #include <net/pkt_sched.h>
41 #include "hyperv_net.h"
44 #define RING_SIZE_MIN 64
45 static int ring_size = 128;
46 module_param(ring_size, int, S_IRUGO);
47 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
49 static int max_num_vrss_chns = 8;
51 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
52 NETIF_MSG_LINK | NETIF_MSG_IFUP |
53 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
56 static int debug = -1;
57 module_param(debug, int, S_IRUGO);
58 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
60 static void do_set_multicast(struct work_struct *w)
62 struct net_device_context *ndevctx =
63 container_of(w, struct net_device_context, work);
64 struct netvsc_device *nvdev;
65 struct rndis_device *rdev;
67 nvdev = hv_get_drvdata(ndevctx->device_ctx);
68 if (nvdev == NULL || nvdev->ndev == NULL)
71 rdev = nvdev->extension;
75 if (nvdev->ndev->flags & IFF_PROMISC)
76 rndis_filter_set_packet_filter(rdev,
77 NDIS_PACKET_TYPE_PROMISCUOUS);
79 rndis_filter_set_packet_filter(rdev,
80 NDIS_PACKET_TYPE_BROADCAST |
81 NDIS_PACKET_TYPE_ALL_MULTICAST |
82 NDIS_PACKET_TYPE_DIRECTED);
85 static void netvsc_set_multicast_list(struct net_device *net)
87 struct net_device_context *net_device_ctx = netdev_priv(net);
89 schedule_work(&net_device_ctx->work);
92 static int netvsc_open(struct net_device *net)
94 struct net_device_context *net_device_ctx = netdev_priv(net);
95 struct hv_device *device_obj = net_device_ctx->device_ctx;
96 struct netvsc_device *nvdev;
97 struct rndis_device *rdev;
100 netif_carrier_off(net);
102 /* Open up the device */
103 ret = rndis_filter_open(device_obj);
105 netdev_err(net, "unable to open device (ret %d).\n", ret);
109 netif_tx_start_all_queues(net);
111 nvdev = hv_get_drvdata(device_obj);
112 rdev = nvdev->extension;
113 if (!rdev->link_state)
114 netif_carrier_on(net);
119 static int netvsc_close(struct net_device *net)
121 struct net_device_context *net_device_ctx = netdev_priv(net);
122 struct hv_device *device_obj = net_device_ctx->device_ctx;
125 netif_tx_disable(net);
127 /* Make sure netvsc_set_multicast_list doesn't re-enable filter! */
128 cancel_work_sync(&net_device_ctx->work);
129 ret = rndis_filter_close(device_obj);
131 netdev_err(net, "unable to close device (ret %d).\n", ret);
136 static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
139 struct rndis_packet *rndis_pkt;
140 struct rndis_per_packet_info *ppi;
142 rndis_pkt = &msg->msg.pkt;
143 rndis_pkt->data_offset += ppi_size;
145 ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
146 rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
148 ppi->size = ppi_size;
149 ppi->type = pkt_type;
150 ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
152 rndis_pkt->per_pkt_info_len += ppi_size;
166 /* Toeplitz hash function
167 * data: network byte order
168 * return: host byte order
170 static u32 comp_hash(u8 *key, int klen, void *data, int dlen)
179 subk.ka = ntohl(*(u32 *)key);
181 for (i = 0; i < dlen; i++) {
182 subk.kb = key[k_next];
183 k_next = (k_next + 1) % klen;
184 dt = ((u8 *)data)[i];
185 for (j = 0; j < 8; j++) {
196 static bool netvsc_set_hash(u32 *hash, struct sk_buff *skb)
198 struct flow_keys flow;
201 if (!skb_flow_dissect_flow_keys(skb, &flow) ||
202 !(flow.basic.n_proto == htons(ETH_P_IP) ||
203 flow.basic.n_proto == htons(ETH_P_IPV6)))
206 if (flow.basic.ip_proto == IPPROTO_TCP)
211 *hash = comp_hash(netvsc_hash_key, HASH_KEYLEN, &flow, data_len);
216 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
217 void *accel_priv, select_queue_fallback_t fallback)
219 struct net_device_context *net_device_ctx = netdev_priv(ndev);
220 struct hv_device *hdev = net_device_ctx->device_ctx;
221 struct netvsc_device *nvsc_dev = hv_get_drvdata(hdev);
225 if (nvsc_dev == NULL || ndev->real_num_tx_queues <= 1)
228 if (netvsc_set_hash(&hash, skb)) {
229 q_idx = nvsc_dev->send_table[hash % VRSS_SEND_TAB_SIZE] %
230 ndev->real_num_tx_queues;
231 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
237 void netvsc_xmit_completion(void *context)
239 struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
240 struct sk_buff *skb = (struct sk_buff *)
241 (unsigned long)packet->send_completion_tid;
244 dev_kfree_skb_any(skb);
247 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
248 struct hv_page_buffer *pb)
252 /* Deal with compund pages by ignoring unused part
255 page += (offset >> PAGE_SHIFT);
256 offset &= ~PAGE_MASK;
261 bytes = PAGE_SIZE - offset;
264 pb[j].pfn = page_to_pfn(page);
265 pb[j].offset = offset;
271 if (offset == PAGE_SIZE && len) {
281 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
282 struct hv_netvsc_packet *packet)
284 struct hv_page_buffer *pb = packet->page_buf;
286 char *data = skb->data;
287 int frags = skb_shinfo(skb)->nr_frags;
290 /* The packet is laid out thus:
291 * 1. hdr: RNDIS header and PPI
293 * 3. skb fragment data
296 slots_used += fill_pg_buf(virt_to_page(hdr),
298 len, &pb[slots_used]);
300 packet->rmsg_size = len;
301 packet->rmsg_pgcnt = slots_used;
303 slots_used += fill_pg_buf(virt_to_page(data),
304 offset_in_page(data),
305 skb_headlen(skb), &pb[slots_used]);
307 for (i = 0; i < frags; i++) {
308 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
310 slots_used += fill_pg_buf(skb_frag_page(frag),
312 skb_frag_size(frag), &pb[slots_used]);
317 static int count_skb_frag_slots(struct sk_buff *skb)
319 int i, frags = skb_shinfo(skb)->nr_frags;
322 for (i = 0; i < frags; i++) {
323 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
324 unsigned long size = skb_frag_size(frag);
325 unsigned long offset = frag->page_offset;
327 /* Skip unused frames from start of page */
328 offset &= ~PAGE_MASK;
329 pages += PFN_UP(offset + size);
334 static int netvsc_get_slots(struct sk_buff *skb)
336 char *data = skb->data;
337 unsigned int offset = offset_in_page(data);
338 unsigned int len = skb_headlen(skb);
342 slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
343 frag_slots = count_skb_frag_slots(skb);
344 return slots + frag_slots;
347 static u32 get_net_transport_info(struct sk_buff *skb, u32 *trans_off)
349 u32 ret_val = TRANSPORT_INFO_NOT_IP;
351 if ((eth_hdr(skb)->h_proto != htons(ETH_P_IP)) &&
352 (eth_hdr(skb)->h_proto != htons(ETH_P_IPV6))) {
356 *trans_off = skb_transport_offset(skb);
358 if ((eth_hdr(skb)->h_proto == htons(ETH_P_IP))) {
359 struct iphdr *iphdr = ip_hdr(skb);
361 if (iphdr->protocol == IPPROTO_TCP)
362 ret_val = TRANSPORT_INFO_IPV4_TCP;
363 else if (iphdr->protocol == IPPROTO_UDP)
364 ret_val = TRANSPORT_INFO_IPV4_UDP;
366 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
367 ret_val = TRANSPORT_INFO_IPV6_TCP;
368 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
369 ret_val = TRANSPORT_INFO_IPV6_UDP;
376 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
378 struct net_device_context *net_device_ctx = netdev_priv(net);
379 struct hv_netvsc_packet *packet = NULL;
381 unsigned int num_data_pgs;
382 struct rndis_message *rndis_msg;
383 struct rndis_packet *rndis_pkt;
387 struct rndis_per_packet_info *ppi;
388 struct ndis_tcp_ip_checksum_info *csum_info;
389 struct ndis_tcp_lso_info *lso_info;
395 struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
396 struct netvsc_stats *tx_stats = this_cpu_ptr(net_device_ctx->tx_stats);
398 /* We will atmost need two pages to describe the rndis
399 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
400 * of pages in a single packet. If skb is scattered around
401 * more pages we try linearizing it.
405 skb_length = skb->len;
406 num_data_pgs = netvsc_get_slots(skb) + 2;
407 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT && linear) {
408 net_alert_ratelimited("packet too big: %u pages (%u bytes)\n",
409 num_data_pgs, skb->len);
412 } else if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
413 if (skb_linearize(skb)) {
414 net_alert_ratelimited("failed to linearize skb\n");
422 pkt_sz = sizeof(struct hv_netvsc_packet) + RNDIS_AND_PPI_SIZE;
424 ret = skb_cow_head(skb, pkt_sz);
426 netdev_err(net, "unable to alloc hv_netvsc_packet\n");
430 /* Use the headroom for building up the packet */
431 packet = (struct hv_netvsc_packet *)skb->head;
434 packet->xmit_more = skb->xmit_more;
436 packet->vlan_tci = skb->vlan_tci;
437 packet->page_buf = page_buf;
439 packet->q_idx = skb_get_queue_mapping(skb);
441 packet->is_data_pkt = true;
442 packet->total_data_buflen = skb->len;
444 packet->rndis_msg = (struct rndis_message *)((unsigned long)packet +
445 sizeof(struct hv_netvsc_packet));
447 memset(packet->rndis_msg, 0, RNDIS_AND_PPI_SIZE);
449 /* Set the completion routine */
450 packet->send_completion = netvsc_xmit_completion;
451 packet->send_completion_ctx = packet;
452 packet->send_completion_tid = (unsigned long)skb;
454 isvlan = packet->vlan_tci & VLAN_TAG_PRESENT;
456 /* Add the rndis header */
457 rndis_msg = packet->rndis_msg;
458 rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
459 rndis_msg->msg_len = packet->total_data_buflen;
460 rndis_pkt = &rndis_msg->msg.pkt;
461 rndis_pkt->data_offset = sizeof(struct rndis_packet);
462 rndis_pkt->data_len = packet->total_data_buflen;
463 rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
465 rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
467 hash = skb_get_hash_raw(skb);
468 if (hash != 0 && net->real_num_tx_queues > 1) {
469 rndis_msg_size += NDIS_HASH_PPI_SIZE;
470 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
472 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
476 struct ndis_pkt_8021q_info *vlan;
478 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
479 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
481 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
483 vlan->vlanid = packet->vlan_tci & VLAN_VID_MASK;
484 vlan->pri = (packet->vlan_tci & VLAN_PRIO_MASK) >>
488 net_trans_info = get_net_transport_info(skb, &hdr_offset);
489 if (net_trans_info == TRANSPORT_INFO_NOT_IP)
493 * Setup the sendside checksum offload only if this is not a
499 if ((skb->ip_summed == CHECKSUM_NONE) ||
500 (skb->ip_summed == CHECKSUM_UNNECESSARY))
503 rndis_msg_size += NDIS_CSUM_PPI_SIZE;
504 ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
505 TCPIP_CHKSUM_PKTINFO);
507 csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
510 if (net_trans_info & (INFO_IPV4 << 16))
511 csum_info->transmit.is_ipv4 = 1;
513 csum_info->transmit.is_ipv6 = 1;
515 if (net_trans_info & INFO_TCP) {
516 csum_info->transmit.tcp_checksum = 1;
517 csum_info->transmit.tcp_header_offset = hdr_offset;
518 } else if (net_trans_info & INFO_UDP) {
519 /* UDP checksum offload is not supported on ws2008r2.
520 * Furthermore, on ws2012 and ws2012r2, there are some
521 * issues with udp checksum offload from Linux guests.
522 * (these are host issues).
523 * For now compute the checksum here.
528 ret = skb_cow_head(skb, 0);
533 udp_len = ntohs(uh->len);
535 uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr,
537 udp_len, IPPROTO_UDP,
538 csum_partial(uh, udp_len, 0));
540 uh->check = CSUM_MANGLED_0;
542 csum_info->transmit.udp_checksum = 0;
547 rndis_msg_size += NDIS_LSO_PPI_SIZE;
548 ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
549 TCP_LARGESEND_PKTINFO);
551 lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
554 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
555 if (net_trans_info & (INFO_IPV4 << 16)) {
556 lso_info->lso_v2_transmit.ip_version =
557 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
558 ip_hdr(skb)->tot_len = 0;
559 ip_hdr(skb)->check = 0;
560 tcp_hdr(skb)->check =
561 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
562 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
564 lso_info->lso_v2_transmit.ip_version =
565 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
566 ipv6_hdr(skb)->payload_len = 0;
567 tcp_hdr(skb)->check =
568 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
569 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
571 lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
572 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
575 /* Start filling in the page buffers with the rndis hdr */
576 rndis_msg->msg_len += rndis_msg_size;
577 packet->total_data_buflen = rndis_msg->msg_len;
578 packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
581 ret = netvsc_send(net_device_ctx->device_ctx, packet);
585 u64_stats_update_begin(&tx_stats->syncp);
587 tx_stats->bytes += skb_length;
588 u64_stats_update_end(&tx_stats->syncp);
590 if (ret != -EAGAIN) {
591 dev_kfree_skb_any(skb);
592 net->stats.tx_dropped++;
596 return (ret == -EAGAIN) ? NETDEV_TX_BUSY : NETDEV_TX_OK;
600 * netvsc_linkstatus_callback - Link up/down notification
602 void netvsc_linkstatus_callback(struct hv_device *device_obj,
603 struct rndis_message *resp)
605 struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
606 struct net_device *net;
607 struct net_device_context *ndev_ctx;
608 struct netvsc_device *net_device;
609 struct rndis_device *rdev;
611 net_device = hv_get_drvdata(device_obj);
612 rdev = net_device->extension;
614 switch (indicate->status) {
615 case RNDIS_STATUS_MEDIA_CONNECT:
616 rdev->link_state = false;
618 case RNDIS_STATUS_MEDIA_DISCONNECT:
619 rdev->link_state = true;
621 case RNDIS_STATUS_NETWORK_CHANGE:
622 rdev->link_change = true;
628 net = net_device->ndev;
630 if (!net || net->reg_state != NETREG_REGISTERED)
633 ndev_ctx = netdev_priv(net);
634 if (!rdev->link_state) {
635 schedule_delayed_work(&ndev_ctx->dwork, 0);
636 schedule_delayed_work(&ndev_ctx->dwork, msecs_to_jiffies(20));
638 schedule_delayed_work(&ndev_ctx->dwork, 0);
643 * netvsc_recv_callback - Callback when we receive a packet from the
644 * "wire" on the specified device.
646 int netvsc_recv_callback(struct hv_device *device_obj,
647 struct hv_netvsc_packet *packet,
648 struct ndis_tcp_ip_checksum_info *csum_info)
650 struct net_device *net;
651 struct net_device_context *net_device_ctx;
653 struct netvsc_stats *rx_stats;
655 net = ((struct netvsc_device *)hv_get_drvdata(device_obj))->ndev;
656 if (!net || net->reg_state != NETREG_REGISTERED) {
657 packet->status = NVSP_STAT_FAIL;
660 net_device_ctx = netdev_priv(net);
661 rx_stats = this_cpu_ptr(net_device_ctx->rx_stats);
663 /* Allocate a skb - TODO direct I/O to pages? */
664 skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
665 if (unlikely(!skb)) {
666 ++net->stats.rx_dropped;
667 packet->status = NVSP_STAT_FAIL;
672 * Copy to skb. This copy is needed here since the memory pointed by
673 * hv_netvsc_packet cannot be deallocated
675 memcpy(skb_put(skb, packet->total_data_buflen), packet->data,
676 packet->total_data_buflen);
678 skb->protocol = eth_type_trans(skb, net);
680 /* We only look at the IP checksum here.
681 * Should we be dropping the packet if checksum
682 * failed? How do we deal with other checksums - TCP/UDP?
684 if (csum_info->receive.ip_checksum_succeeded)
685 skb->ip_summed = CHECKSUM_UNNECESSARY;
687 skb->ip_summed = CHECKSUM_NONE;
690 if (packet->vlan_tci & VLAN_TAG_PRESENT)
691 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
694 skb_record_rx_queue(skb, packet->channel->
695 offermsg.offer.sub_channel_index);
697 u64_stats_update_begin(&rx_stats->syncp);
699 rx_stats->bytes += packet->total_data_buflen;
700 u64_stats_update_end(&rx_stats->syncp);
703 * Pass the skb back up. Network stack will deallocate the skb when it
712 static void netvsc_get_drvinfo(struct net_device *net,
713 struct ethtool_drvinfo *info)
715 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
716 strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
719 static void netvsc_get_channels(struct net_device *net,
720 struct ethtool_channels *channel)
722 struct net_device_context *net_device_ctx = netdev_priv(net);
723 struct hv_device *dev = net_device_ctx->device_ctx;
724 struct netvsc_device *nvdev = hv_get_drvdata(dev);
727 channel->max_combined = nvdev->max_chn;
728 channel->combined_count = nvdev->num_chn;
732 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
734 struct net_device_context *ndevctx = netdev_priv(ndev);
735 struct hv_device *hdev = ndevctx->device_ctx;
736 struct netvsc_device *nvdev = hv_get_drvdata(hdev);
737 struct netvsc_device_info device_info;
738 int limit = ETH_DATA_LEN;
740 if (nvdev == NULL || nvdev->destroy)
743 if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
744 limit = NETVSC_MTU - ETH_HLEN;
746 /* Hyper-V hosts don't support MTU < ETH_DATA_LEN (1500) */
747 if (mtu < ETH_DATA_LEN || mtu > limit)
750 nvdev->start_remove = true;
751 cancel_work_sync(&ndevctx->work);
752 netif_tx_disable(ndev);
753 rndis_filter_device_remove(hdev);
757 ndevctx->device_ctx = hdev;
758 hv_set_drvdata(hdev, ndev);
759 device_info.ring_size = ring_size;
760 device_info.max_num_vrss_chns = max_num_vrss_chns;
761 rndis_filter_device_add(hdev, &device_info);
762 netif_tx_wake_all_queues(ndev);
767 static struct rtnl_link_stats64 *netvsc_get_stats64(struct net_device *net,
768 struct rtnl_link_stats64 *t)
770 struct net_device_context *ndev_ctx = netdev_priv(net);
773 for_each_possible_cpu(cpu) {
774 struct netvsc_stats *tx_stats = per_cpu_ptr(ndev_ctx->tx_stats,
776 struct netvsc_stats *rx_stats = per_cpu_ptr(ndev_ctx->rx_stats,
778 u64 tx_packets, tx_bytes, rx_packets, rx_bytes;
782 start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
783 tx_packets = tx_stats->packets;
784 tx_bytes = tx_stats->bytes;
785 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
788 start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
789 rx_packets = rx_stats->packets;
790 rx_bytes = rx_stats->bytes;
791 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
793 t->tx_bytes += tx_bytes;
794 t->tx_packets += tx_packets;
795 t->rx_bytes += rx_bytes;
796 t->rx_packets += rx_packets;
799 t->tx_dropped = net->stats.tx_dropped;
800 t->tx_errors = net->stats.tx_dropped;
802 t->rx_dropped = net->stats.rx_dropped;
803 t->rx_errors = net->stats.rx_errors;
808 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
810 struct net_device_context *ndevctx = netdev_priv(ndev);
811 struct hv_device *hdev = ndevctx->device_ctx;
812 struct sockaddr *addr = p;
813 char save_adr[ETH_ALEN];
814 unsigned char save_aatype;
817 memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
818 save_aatype = ndev->addr_assign_type;
820 err = eth_mac_addr(ndev, p);
824 err = rndis_filter_set_device_mac(hdev, addr->sa_data);
826 /* roll back to saved MAC */
827 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
828 ndev->addr_assign_type = save_aatype;
834 #ifdef CONFIG_NET_POLL_CONTROLLER
835 static void netvsc_poll_controller(struct net_device *net)
837 /* As netvsc_start_xmit() works synchronous we don't have to
838 * trigger anything here.
843 static const struct ethtool_ops ethtool_ops = {
844 .get_drvinfo = netvsc_get_drvinfo,
845 .get_link = ethtool_op_get_link,
846 .get_channels = netvsc_get_channels,
849 static const struct net_device_ops device_ops = {
850 .ndo_open = netvsc_open,
851 .ndo_stop = netvsc_close,
852 .ndo_start_xmit = netvsc_start_xmit,
853 .ndo_set_rx_mode = netvsc_set_multicast_list,
854 .ndo_change_mtu = netvsc_change_mtu,
855 .ndo_validate_addr = eth_validate_addr,
856 .ndo_set_mac_address = netvsc_set_mac_addr,
857 .ndo_select_queue = netvsc_select_queue,
858 .ndo_get_stats64 = netvsc_get_stats64,
859 #ifdef CONFIG_NET_POLL_CONTROLLER
860 .ndo_poll_controller = netvsc_poll_controller,
865 * Send GARP packet to network peers after migrations.
866 * After Quick Migration, the network is not immediately operational in the
867 * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
868 * another netif_notify_peers() into a delayed work, otherwise GARP packet
869 * will not be sent after quick migration, and cause network disconnection.
870 * Also, we update the carrier status here.
872 static void netvsc_link_change(struct work_struct *w)
874 struct net_device_context *ndev_ctx;
875 struct net_device *net;
876 struct netvsc_device *net_device;
877 struct rndis_device *rdev;
878 bool notify, refresh = false;
879 char *argv[] = { "/etc/init.d/network", "restart", NULL };
880 char *envp[] = { "HOME=/", "PATH=/sbin:/usr/sbin:/bin:/usr/bin", NULL };
884 ndev_ctx = container_of(w, struct net_device_context, dwork.work);
885 net_device = hv_get_drvdata(ndev_ctx->device_ctx);
886 rdev = net_device->extension;
887 net = net_device->ndev;
889 if (rdev->link_state) {
890 netif_carrier_off(net);
893 netif_carrier_on(net);
895 if (rdev->link_change) {
896 rdev->link_change = false;
904 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
907 netdev_notify_peers(net);
910 static void netvsc_free_netdev(struct net_device *netdev)
912 struct net_device_context *net_device_ctx = netdev_priv(netdev);
914 free_percpu(net_device_ctx->tx_stats);
915 free_percpu(net_device_ctx->rx_stats);
919 static int netvsc_probe(struct hv_device *dev,
920 const struct hv_vmbus_device_id *dev_id)
922 struct net_device *net = NULL;
923 struct net_device_context *net_device_ctx;
924 struct netvsc_device_info device_info;
925 struct netvsc_device *nvdev;
927 u32 max_needed_headroom;
929 net = alloc_etherdev_mq(sizeof(struct net_device_context),
934 max_needed_headroom = sizeof(struct hv_netvsc_packet) +
937 netif_carrier_off(net);
939 net_device_ctx = netdev_priv(net);
940 net_device_ctx->device_ctx = dev;
941 net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
942 if (netif_msg_probe(net_device_ctx))
943 netdev_dbg(net, "netvsc msg_enable: %d\n",
944 net_device_ctx->msg_enable);
946 net_device_ctx->tx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
947 if (!net_device_ctx->tx_stats) {
951 net_device_ctx->rx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
952 if (!net_device_ctx->rx_stats) {
953 free_percpu(net_device_ctx->tx_stats);
958 hv_set_drvdata(dev, net);
959 INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
960 INIT_WORK(&net_device_ctx->work, do_set_multicast);
962 net->netdev_ops = &device_ops;
964 net->hw_features = NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_IP_CSUM |
966 net->features = NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_SG | NETIF_F_RXCSUM |
967 NETIF_F_IP_CSUM | NETIF_F_TSO;
969 net->ethtool_ops = ðtool_ops;
970 SET_NETDEV_DEV(net, &dev->device);
973 * Request additional head room in the skb.
974 * We will use this space to build the rndis
975 * heaser and other state we need to maintain.
977 net->needed_headroom = max_needed_headroom;
979 /* Notify the netvsc driver of the new device */
980 device_info.ring_size = ring_size;
981 device_info.max_num_vrss_chns = max_num_vrss_chns;
982 ret = rndis_filter_device_add(dev, &device_info);
984 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
985 netvsc_free_netdev(net);
986 hv_set_drvdata(dev, NULL);
989 memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
991 nvdev = hv_get_drvdata(dev);
992 netif_set_real_num_tx_queues(net, nvdev->num_chn);
993 netif_set_real_num_rx_queues(net, nvdev->num_chn);
995 ret = register_netdev(net);
997 pr_err("Unable to register netdev.\n");
998 rndis_filter_device_remove(dev);
999 netvsc_free_netdev(net);
1001 schedule_delayed_work(&net_device_ctx->dwork, 0);
1007 static int netvsc_remove(struct hv_device *dev)
1009 struct net_device *net;
1010 struct net_device_context *ndev_ctx;
1011 struct netvsc_device *net_device;
1013 net_device = hv_get_drvdata(dev);
1014 net = net_device->ndev;
1017 dev_err(&dev->device, "No net device to remove\n");
1021 net_device->start_remove = true;
1023 ndev_ctx = netdev_priv(net);
1024 cancel_delayed_work_sync(&ndev_ctx->dwork);
1025 cancel_work_sync(&ndev_ctx->work);
1027 /* Stop outbound asap */
1028 netif_tx_disable(net);
1030 unregister_netdev(net);
1033 * Call to the vsc driver to let it know that the device is being
1036 rndis_filter_device_remove(dev);
1038 netvsc_free_netdev(net);
1042 static const struct hv_vmbus_device_id id_table[] = {
1048 MODULE_DEVICE_TABLE(vmbus, id_table);
1050 /* The one and only one */
1051 static struct hv_driver netvsc_drv = {
1052 .name = KBUILD_MODNAME,
1053 .id_table = id_table,
1054 .probe = netvsc_probe,
1055 .remove = netvsc_remove,
1058 static void __exit netvsc_drv_exit(void)
1060 vmbus_driver_unregister(&netvsc_drv);
1063 static int __init netvsc_drv_init(void)
1065 if (ring_size < RING_SIZE_MIN) {
1066 ring_size = RING_SIZE_MIN;
1067 pr_info("Increased ring_size to %d (min allowed)\n",
1070 return vmbus_driver_register(&netvsc_drv);
1073 MODULE_LICENSE("GPL");
1074 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1076 module_init(netvsc_drv_init);
1077 module_exit(netvsc_drv_exit);